Summary of formulas:

xj column in optimal constraints = B-1aj RHS of optimal tableaus constraints = B-1b Coefficient of NBV in optimal Row 0: j = cBVB-1aj-cj Slack variable, si : ith element of cBV B-1 1) Excess variable, variable ei : - (ith element of cBV B-1 Artificial variable, ai : (ith element of cBV B-1) + M (Max problem) RHS of optimal Row 0 = cBV B-1b
1

We must find B-1 to compute all parts of optimal tableau We must find cBV B-1 to compute the optimal tableau Row 0.

Example p 1: ( (Refer to Example p 4 Chapter p 3) ) For the following LP, the optimal basis is }. Compute p the optimal p tableau. BV = { x2,s2 } min z = 2x1 - 5x2 s.t 3x1+8x2 12 2x1+3x2 6 x1,x2 0 Solution 1: The standard form min z = 2x1 - 5x2 s.t 3x1+8x2 +s1 =12 +s2 = 6 2x1+3x2 x1,x2 ,s1,s2 0
3

Solution 1: The standard form min z = - 5x2+s2+ 2x1+s1 s.t t 8x 8 2 + 3x 3 1+s1 =12 12 3x2 + s2 + 2x1 =6 x1,x2 ,s1,s2 0

x x z = [ 5 0] 2 + [ 2 0] 1 s2 s1 8 0 x2 3 1 x1 12 3 1 s + 2 0 s = 6 2 1 x2 x1 s 0, s 0 2 1

x cBV = [ 5 0] , xBv = 2 , cBNV = [ 2 0] , s2 x 8 0 3 1 12 , = , = xNBV = 1 , B = N b 2 0 6 3 1 s1

Step 1: Find cBV and B-1 cBV=[-5 0]

1 0 8 8 0 -1 B = so by using Gauss elimination B = 3 1 3 1 8

4

 Step 2: Determine the optimal constraints

Row 0 1 2 BV z x2 s2 z 1 0 0 x2 0 1 0 s2 0 0 1 x1 s1 RHS

LHS=formula B-1aj
1 8 1 x1 = B a1 = 3 8 1 8 1 s1 = B as1 = 3 8 3 0 3 8 = 2 7 1 8 1 0 1 8 = 0 3 1 8

RHS=formula B-1b

1 3 0 8 12 2 = 3 6 3 1 8 2
5

Thus, the constraints of the optimal tableau are

3 x2 8 s + 7 2 8

1 3 8 x1 2 = 3 s1 3 2 8

 Step 3: Determine the optimal Row 0

LHS (coefficient for NBV) c j = cBV B 1 a j c j 31 5 3 c1 = 0 2 = 8 8 2 5 5 1 cs1 = 0 0 = 8 8 0

cBV B 1

1 0 8 5 = [ 5 0] 0 = 3 1 8 8
RHS

15 5 12 cBV B b = 0 = 8 8 6
1

Row 0 1 2

BV z x2 s2

z 1 0 0

x2 0 1 0

s2 0 0 1

x1

s1

RHS

3/8 7/8

1/8 -3/8

3/2 3/2
7

 Remember Remember, the BVs will have zero coefficient in Row 0. Thus, the complete optimal tableau z (31/8)x1 (5/8)x2 = -15/2 (3/8) 1+ x2 +s1 = 3/2 (3/8)x (7/8)x1-(3/8)x2+s2= 3/2

4.3 Sensitivity Analysis (3)

A simplex tableau (max problem) for a set of BV is optimal iff Each constraint has nonnegative RHS Each variable has a nonnegative coefficient in Row 0. Procedure to determine if any change in LP will cause BV to be no longer optimal 1) Using formula in Sensitivity Analysis (2), determine how changes in LP parameter change the RHS and Row 0 2) If each variable in Row 0 has nonnegative coefficient and each constraint has nonnegative RHS, BV is still optimal Otherwise optimal. Otherwise, BV is no longer optimal optimal. 3) If BV is no longer optimal, find the new optimal solution. Recreate the entire tableau for BV and continuing the simplex algorithm. algorithm
9

 If any constraint t i t has h a negative ti RHS, RHS BV is now an infeasible basis. If a variable in Row 0 may y have a negative g coefficient, BV is now a suboptimal basis.

Three types of changes in an LP parameter ) Changing g g the objective j function coefficient of a 1) nonbasic variable. 2) Changing the objective function coefficient of a basic variable. 3) Changing the right hand side (RHS) of a constraint.
10

4.3.1 Changing the Objective Function Coefficient of a Nonbasic Variable

Reconsider Dakota Problem. (1) is original problem in standard form.

and the optimal tableau

11

ROW 0 1 2 3

BV z s1 x3 x1

z 1 0 0 0

x1 0 0 0 1

x2 5 -2 -2 1.25

x3 0 0 1 0

s1 0 1 0 0

s2 10 2 2 -0.5

s3 10 -8 -4 1.5

RHS 280 24 8 2

12

xBV

s1 , x = 3 x1

xNBV

x2 , s = 2 s3

2 8 1 B1 = 0 2 4 0 1/ 2 3 / 2

For BV has to follow the order from optimal tableau

F NBV no need For di in order d

The only nonbasic decision variable is x2. Coefficient of x2 of objective function is currently c2 =30

Question: For what value of c2 would BV remain optimal?

13

Question: F what For h t value l of f c2 would ld BV remain i optimal? ti l? Suppose we change c2 from 30 to 30 30+ so B remain unchange. Step 1: Determine whether RHS of constraint is nonnegative. Check Formula: B-1b Does B-1 changed? NO Does b changed? NO Therefore RHS of constraints have nonnegative Therefore, values. BV is still feasible (remain unchanged)
14

 Step p 2: Determine the coefficient of NBV in optimal Row 0 Check Formula: j=cBVB-1aj-cjj. BV remain optimal if j 0 p if j < 0 BV will be suboptimal The only variable will be change is

c2 = cBV B a2 c2 6 = [ 0 10 10] 2 ( 30 + ) = 5 .5 1.5

15

 BV will remain optimal if 5-0 5 If the coefficient of x2 is decreased or increased by 5 or less, (in other word, c2 30+5=35) BV will remains optimal. If c2>35 35 , BV will ill b be suboptimal b ti l ( (no l longer optimal). This is because j < 0. We need to find the new optimal solution solution. We can increase z by making x2 as basic variables
16

For example: c2=40

6 40 = 5 c2 = cBV B 1 a2 c2 = [ 0 10 10] 2 1.5 Final (suboptimal) Tablaeu
ROW 0 1 2 3 BV z s1 x3 x1 z 1 0 0 0 x1 0 0 0 1 x2 -5 5 -2 -2 1.25 x3 0 0 1 0 s1 0 1 0 0 s2 10 2 2 -0.5 s3 10 -8 -4 1.5 RHS 280 24 8 2
17

Pivot element

Optimal Tablaeu

ROW 0 1 2 3

BV z s1 x3 x2

z 1 0 0 0

x1 4 1.6 1.6 0.8

x2 0 0 0 1

x3 0 0 1 0

s1 0 1 0 0

s2 8 1.2 1.2 -0.4

s3 16 -5.6 -1.6 1.2

RHS 280 27.2 11.2 1.6

z1=280 when x1=0, =0 x2=1.6, =1 6 x3=11.2 =11 2

18

4.3.2 Changing the Objective Function Coefficient of a Basic Variable

Suppose that c1 is change from 60 to 60 60+ cBV will be change to [0 20 60+] Step 1: Determine whether RHS of constraint is nonnegative. Check Formula: B-1b Does B-1 changed? NO Does b changed? NO Therefore, RHS of constraints have nonnegative values. BV is still feasible (remain unchanged)
19

 Step 2: Determine the coefficient of NBV in optimal R Row 0 Check Formula: j=cBVB-1aj-cj. BV remain optimal p if j 0 BV will be suboptimal if j < 0
2 8 1 0 = 0 10 0.5 10 + 0.5 cBV B 1 = [ 0 20 60 + ] 2 4 [ ] 0 0.5 1.5 s1 , x3 , x1 are BV, BV th their i coefficient ffi i t in i optimal ti l tablaeu t bl Row R 0 must t still till be b 0 6 30 = 5 + 1.25 c2 = cBV B 1 a2 c2 = [ 0 10 0.5 10 + 0.5 ] 2 1.5 cs 2 = cBV B 1 as 2 cs 2 = 2th element of cBV B 1 = 10 0.5 cs 3 = cBV B 1 as 3 cs 3 = 3th element of cBV B 1 = 10 + 0.5
20

BV remain optimal iff c2 0, cs 2 0, cs 3 0, 5 + 1.25 0 10 0.5 0 10 + 0.5 0 4 20 -20/3

Therefore, current basis remain optimal for -4 20 or 56=60-4 c1 60+20=80

21

When the current basis is no longer optimal

Let c1=100(or =40)
s1 , x3 , x1 are BV, their coefficient in optimal tablaeu Row 0 must still be 0 cs1 = c3 = c1 = 0 c2 = 5 + 1.25 = 55 cs 2 = 10 0.5 = 10 cs 3 = 10 + 0.5 = 70 48 = 360 RHS of row 0, cBV B 1b = [ 0 10 70] 20 8
22

Final (suboptimal tableau)

ROW 0 1 2 3

BV z s1 x3 x1

z 1 0 0 0

x1 0 0 0 1

x2 55 -2 2 -2 1.25

x3 0 0 1 0

s1 0 1 0 0

s2 -10 2 2 -0.5

s3 70 -8 8 -4 1.5

RHS 360 24 8 2

Pivot element
23

Optimal tableau

ROW 0 1 2 3

BV z s1 s3 x1

z 1 0 0 0

x1 0 0 0 1

x2 45 0 -1

x3 5 -1 0.5

s1 0 1 0 0

s2 0 0 1 0

s3 50 -4 -2 0.5

RHS 400 16 4 4

0.75 0.25

z=400, when x1=4, x2=0, x3=0

24

4.3.3 Changing the RHS of a Constraint

Changing the RHS cannot cause the current basic to become suboptimal. As long as the RHS remains nonnegative, nonnegative current basis remains feasible and optimal. Suppose pp we change g b2 from 20 to 20+ Step 1: Determine whether RHS of constraint is nonnegative. 2 8 48 1 Check Formula: B-1 b 20 + B 1b = 0 2 4 -1 Does B changed? NO 0 0.5 1.5 8 Does b changed? YES 24 + 2 Therefore, = 8 + 2 RHS of constraints 2 0.5 0 5

25

 Current basis will remain optimal iff

24 + 2 0 8 + 2 0 2 0.5 0 12 4 4

Therefore, current basis remain optimal for -44 or 16 =20-4 b220+4 =24
26

Effect on decision variables and z

If 16 b224, , the current basis remain optimal p (s1,x3,x1) but the values of decision variables and z change. Suppose S b2=22 22 (where ( h 20+2 20+2, D=2) 2) so
2 8 48 28 s1 1 x = B 1b = 0 20 + 2 = 12 2 4 3 x1 0 0.5 1.5 8 1 48 = 300 z = cBV B 1b = [ 0 10 10] 20 2 + 8
27

When the current basis is no longer optimal:

Suppose b2=30 (or =10) RHS of optimal constraints:
2 8 48 44 s1 1 x = B 1b = 0 20 + 10 = 22 2 4 3 05 1 1.5 5 x1 0 0.5 8 3 Since x1 = 3, BV is no longer feasible RHS of row 0 48 = 380 z = cBV B 1b = [ 0 10 10] 20 10 + 8

We will use dual simplex algorithm to solve LPs when the initial tableau has one or more ve RHS and each variable in Row 0 has nonnegative coefficient.
28